Memory managment with classes Best practice to allocate/deallocate object

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everybody,

If we take a simple class as follows:

class ClassTest

{

     int size;

     int *tab;

};

With Cuda, I have not found another way to allocate an object of type ClassTest like this:

// consider an object in the host memory contains data

      ClassTest *h_object;

// temporary pointer on ClassTest object

      ClassTest *tmp_test = new ClassTest();

      tmp_test->size = h_object->size;

      cudaMalloc((void **)&tmp_test->tab, sizeof(int) * tmp_test->size);

      cudaMemcpy(tmp_test->tab, h_object->tab, sizeof(int) * tmp_test->size , cudaMemcpyHostToDevice);

// Pointer on real object cuda

      ClassTest *d_test;

      cudaMalloc((void **)&d_test, sizeof(ClassTest));

      cudaMemcpy(d_test, tmp_test, sizeof(ClassTest), cudaMemcpyHostToDevice);

And this code for deallocate device object :

cudaFree(tmp_test->tab);

      delete tmp_test;

      cudaFree(d_test);

So, I wondered if there was a better way to do this?

Thanks

2

Yet, nobody as never copy object from host to device memory ?

3

I generally avoid data structures on the device that contain pointers for this reason. There isn’t an easier way to do this.

4

Ok, Thanks

And is that the performance is lower with the use of classes like this instead of simple array ?